Prompt
7 minute readSearch Console Prompt: Find Opportunities in Query Data
A reusable Search Console prompt that turns a query and page export into a short action list: low-CTR pages, near-page-one queries and competing pages, each tied to the exact rows.
This Search Console prompt reviews a query and page export and turns it into a short, prioritised action list. It finds pages with impressions but few clicks, queries sitting just off page one and pages that compete for the same query. Every finding cites the exact rows behind it, so you can check it in seconds.
The page explains each variable, gives the full prompt and shows one real run on an invented export. It also lists the checks that catch arithmetic mistakes. The model only reads the numbers you paste; it cannot see your account or know why a number changed.
What this Search Console prompt does
The prompt sorts rows into three common opportunity types: high impressions with low CTR, average positions roughly between 8 and 20, and one query spread across several pages. For each finding it suggests one next step, such as a title review, a content update or a canonical decision.
It uses only the columns you provide: query, page, clicks, impressions, CTR and position. It must not invent search volume, traffic forecasts or reasons for a change. Where the data cannot answer a question, it says so.
When to use it
Use it for a monthly review of one site section, or when you need a quick first read before a deeper analysis. It suits exports of up to a few hundred rows.
- You want a shortlist of pages to refresh this month.
- You suspect two pages compete for the same query.
- You need a plain summary of an export for a colleague.
However, for large exports or exact calculations, use a spreadsheet or a script and give the model the summary. Models make arithmetic mistakes, so keep the maths in a tool you trust.
Inputs and variables
Replace every bracketed variable before you run the prompt. Each one shapes what counts as an opportunity.
- [SITE_CONTEXT]: one sentence on the site and the section you are reviewing.
- [DATE_RANGE]: the exact range of the export, for example "October 1 to October 28, 2026".
- [EXPORT_ROWS]: rows as query, page, clicks, impressions, CTR, position. Export them from the Performance report with the Queries and Pages dimensions.
- [EXCLUSIONS]: queries or traffic to leave out, such as branded searches or your own test traffic. Write "none" if there are none.
Search Console data is your company's information. Check your organisation's rules before you paste it into a public AI tool, and never include personal data.
Reusable prompt — version 1.0
Copy the prompt below as it is. It asks the model to quote the rows behind every finding, which makes errors easy to spot.
You are helping an SEO analyst review a Google Search Console export for [SITE_CONTEXT].
<task>
Find the most useful opportunities in <rows> and propose one next step for each.
</task>
<context>
Date range: [DATE_RANGE]
Leave out: [EXCLUSIONS]
Columns: query, page, clicks, impressions, CTR, position (position is an average).
</context>
<rules>
1. Use only the numbers in <rows>. Never invent search volume, traffic forecasts, rankings or dates.
2. Quote the exact rows behind every finding.
3. Look for three types: (a) high impressions with low CTR, (b) average position between about 8 and 20, (c) one query that sends impressions to more than one page.
4. Do not explain why a number changed; the export cannot show causes. Say what to check instead.
5. If two rows conflict or a value looks wrong, list it under "Data questions" rather than using it.
6. Rank findings by impressions, highest first, and keep the list to 10 or fewer.
</rules>
<output_format>
Table: type | page | query | clicks | impressions | CTR | position | next step
List: Data questions
List: What this export cannot tell you
</output_format>
<rows>
[EXPORT_ROWS]
</rows>Rule 4 keeps the model from inventing explanations. Instead of "the page dropped because of an update", it points you to the check that would tell you.
How to use it
Any capable chat model can run this Search Console prompt. Smaller, focused exports give more reliable results than a full-site dump.
- In the Performance report, set the date range and export queries and pages for one section.
- Remove branded queries and known test traffic, or list them in [EXCLUSIONS].
- Fill in the variables, paste the prompt into a new chat and run it.
- Check each quoted row against your export, and recalculate any CTR the model used.
- Turn confirmed findings into tasks with an owner and a review date.
Example input and output
We ran version 1.0 once on Claude Opus 5.5 on November 15, 2026. The export is invented for this example and has 9 rows; the output is trimmed, and trims are marked with "…".
SITE_CONTEXT: Northwind Bikes, repair guides section
DATE_RANGE: October 1 to October 28, 2026
EXCLUSIONS: queries containing "northwind"
EXPORT_ROWS:
disc brake rubbing, /guides/adjust-disc-brakes, 12, 1840, 0.65%, 9.4
how to adjust disc brakes, /guides/adjust-disc-brakes, 41, 960, 4.27%, 6.1
fix flat bike tire, /guides/fix-a-flat-tire, 3, 410, 0.73%, 14.8
fix flat bike tire, /blog/flat-tire-kit, 1, 220, 0.45%, 18.2
…| type | page | query | clicks | impressions | CTR | position | next step |
| low CTR | /guides/adjust-disc-brakes | disc brake rubbing | 12 | 1840 | 0.65% | 9.4 | Check whether the title and opening mention rubbing; the guide covers it. |
| two pages | /guides/fix-a-flat-tire and /blog/flat-tire-kit | fix flat bike tire | 3 and 1 | 410 and 220 | 0.73% and 0.45% | 14.8 and 18.2 | Decide which page should own this query and link the other to it. |
…
Data questions:
- None found in these rows.
What this export cannot tell you:
- Why CTR is low (position, snippet or competing results); compare the live results page.In this run, every quoted number matched the input, and the model did not guess at causes. That single run is an illustration, not a benchmark; always recheck the numbers.
How to check the output
Recheck the findings against your own export before you act on them.
- Rows: every quoted row exists in your export with the same numbers.
- Maths: any CTR or total the model computed is correct when you recalculate it.
- No invention: there is no search volume, forecast or cause that the data cannot show.
- Exclusions: branded and test traffic stayed out of the findings.
- Next steps: each one fits the page and is something you can check or do.
Troubleshooting
These fixes address the bad outputs we see most often.
- The model explains causes anyway: repeat rule 4 at the end of the prompt.
- Numbers in the table differ from the input: paste fewer rows, and ask it to copy numbers without rounding.
- Everything looks like an opportunity: add a minimum, for example "ignore rows with fewer than 100 impressions".
- Branded queries still appear: remove them from the export instead of relying on [EXCLUSIONS].
Limitations
Search Console reports average position and may omit some queries for privacy reasons. Therefore, an export is a partial view, and the model cannot fill the gaps. It also cannot see results pages, so it cannot tell why searchers did not click.
Outputs vary by model, version and input. Treat the table as a draft for a person to verify, not as a report to send unchecked.
Version history
Each version is dated so you can tell which prompt produced an older review.
- 1.0: November 15, 2026. First version, tested once on Claude Opus 5.5.
Frequently asked questions
Quick answers to the questions readers ask most about this topic.
Which AI models does this Search Console prompt work with?
Any capable chat model that follows structured instructions, such as current models from OpenAI, Anthropic or Google. We tested it once on Claude Opus 5.5. Recheck every number, because models can make arithmetic mistakes.
Is it safe to paste Search Console data into an AI tool?
It is your company's data, so check your organisation's rules first. Never include personal data, and prefer tools whose data policies you have reviewed.
How many rows should I paste?
A few hundred at most. For larger exports, summarise the data in a spreadsheet first and give the model the summary, or analyse one site section at a time.
Can the prompt tell me why my traffic dropped?
No. A Search Console export shows what changed, not why. The prompt lists what to check instead, such as the live results page, recent site changes and known search updates.
Sources
These pages support the Search Console details and the prompting approach above.
Keep reading